Documentation

curl -fsS https://rhei.team/install | bash

Requires macOS or Linux (Windows via WSL), git, and Bun or Node 18+. The installer puts the rhei CLI — which is also the local MCP server — on your machine. Nothing about your code is sent anywhere during install.

Rhei is in closed beta: if the public package is not available yet, the installer tells you and points you to app.rhei.team/cli for access.

cd your-repo
rhei init

rhei init maps the repo, builds the local code graph and symbol index, connects your MCP clients, runs a smoke check, and shows local status. No account, no upload — Pro is a separate, explicit opt-in at the end.

rhei init offers to register Rhei with the MCP clients it finds — Claude Code, Codex CLI, Cursor, and other MCP-capable agents. To wire a client manually, run the CLI as a stdio MCP server:

# Claude Code
claude mcp add rhei -- rhei mcp

# any MCP client — stdio server config
{ "command": "rhei", "args": ["mcp"] }

After changing MCP config or updating Rhei, restart or reconnect the client so it picks up the new server. rhei doctor verifies the wiring end to end.

Agents see one consolidated surface of 20 tools instead of a pile of one-off utilities. Grouped by intent:

• Read & navigate

  read_file · file_search · get_file_tree · get_code_structure

  Exact reads, symbol/path-indexed search, repo tree, and code structure — the agent's eyes on the repo.

• Context & goals

  goal · context_builder · workspace_context · manage_selection · graph

  Goal-scoped context packets with receipts: relevant files, slices, symbols, token accounting, and omitted candidates.

• Change & verify

  apply_edits · file_actions · git · validate

  Gated edits with previews and receipts, file operations, working-tree status, and validation evidence.

• State & memory

  index · status · ledger

  Index lifecycle, readiness posture, and the evidence ledger that records what agents actually used.

• Agents & oracle

  agent_run · agent_manage · oracle_send · oracle_utils

  Report-only handoffs to sub-agents and oracle sessions — packaged work in, evidence out.

Edits are gated and receipt-backed: previews by default, apply only when explicit gates allow it, and durable verification after every write.

• Local code graph — files, symbols, and relationships are indexed incrementally on your machine.
• Query routing — each query is classified (symbol, literal, intent) and routed to the matching retrieval strategy.
• Residency — hot, warm, and cold tiers keep quiet code cheap and bring active code forward.
• Workspace memory — the graph learns from what agents actually use, keeping usefulness separate from truth.
• Receipts — every context decision reports what was selected, what was omitted, and why.

The system overview on the home page walks through each stage with measured replay numbers.

Local mode keeps source, the graph index, reports, and memory on your machine. Paid plans only send what your configured source-upload mode allows: never, metadata_only, or selected_slices. Rhei does not train models on customer code by default.

Full contracts: Data Use overview · Security · Privacy Policy.

rhei update        # dry-run: report current vs. latest
rhei update --yes  # apply, then restart your MCP client

Update checks resolve against the published manifest at rhei.team/.well-known/rhei-mcp-update.json — no telemetry required. Updates never touch your source.